Spatial Inference of Vegetation Vulnerability for the Ecological Economical Zoning of Minas Gerais Luis M. T. Carvalho 1 Moisés S. Ribeiro 2, Luciano T.

Spatial Inference of Vegetation Vulnerabilityfor the

Ecological Economical Zoning of Minas Gerais

Luis M. T. Carvalho1

Moisés S. Ribeiro2, Luciano T. de Oliveira1

Thomaz C. A. Oliveira1, Julio N. Louzada3

José R. S. Scolforo1, Antonio D. Oliveira1

1Departamento de Ciências Florestais2Departamento de Engenharia

3Departamento de Biologia


Introduction

ZEE → Zones subject to a certain model of use according to

degrees of natural vulnerability and social potentiality.

ZEE-MG → implemented by the Government of Minas Gerais to

support policy making by means of a statewide diagnosis of

economical, social, ecological and biophysical sustainability.

NATURAL VULNERABILITY → the capacity of resisting or

recovering from impacts caused by human activities.


Introduction

ZEE/MG

Vulnerability Potentiality

Institutional

Productive

Biotic

Physical

Natural

Human

Flora

Fauna

Soils

Erosion

Water

Climate


Objectives

to investigate alternative methods of spatial inference, viz. fuzzy

logic and neural networks for generating maps of vegetation

vulnerability for the State of Minas Gerais, and

to evaluate their suitability to be used instead of weighted

overlay.


Study site and data sets

The study area comprises the whole State of Minas Gerais.

Data compiled and included in the ZEE-MG were structured in a

GIS using the raster data model with a spatial resolution of

270x270m.

Indicators of vegetation vulnerability were derived from a

30x30m resolution land cover map (Scolforo & Carvalho, 2006)

and from priority conservation areas (Drummond et al., 2005)


Flora

Conservation Heterogeneity RelevanceConservation

Priority

Indicators of Vegetation Vulnerability


30m

6 : Regional total

270m

Rocky Field Cerrado stricto sensu Semideciduous Forests

Indicators 1 to 9: Regional Relevance


Grass land Rocky grass land Open savanna

Savanna stricto sensu Savanna woodland Savanna palm land

Deciduous forest Semi deciduous forest Evergreen forest


30m

11

270m

Native Vegetation Others

Indicator 10: Conservation Degree



30m

3

270m

Campo rupestre Cerrado stricto sensu Floresta Estacional Semidecidual

Indicator 11: Spatial Heterogeneity



Conservation Priority classes Vulnerability classes

None Very low

Corridor Low

Potential Medium

High High

Very high, Extreme and Special Very high

Indicator 12: Conservation Priority



Methods

Albers Conic Equal Area Projection (datum SAD-69).

Spatial inference using weighted overlay, fuzzy logic, and neural

networks.

Vulnerability represented by the models outputs were classified

as (1) Very low, (2) Low, (3) Medium, (4) High, and (5) Very high.


Weighted Overlay

Simple and straightforward technique.

Weights represent the importance of each variable, as well as the

importance of each classe according to a given objective.

Allows the inclusion of expert knowledge.

1

2 3

1 1 1 1

1 1

1

2

2

2 2

3 3

3

3

32

2

2 2

2 2

2 2

Peso = 75% Peso = 25%

+ =

Ecological Economical Zoning of Minas GeraisIndicator Indicator weight Class Class weight

Regional relevance 8 Very low 1

Low 6

Medium 10

High 12

Very high 12

Degree of conservation 12 Very low 1

Low 6

Medium 10

High 12

Very high 12

Spatial heterogeneity 4 Very low 1

Low 6

Medium 10

High 12

Very high 12

Conservation priority 12 Very low 1

Low 2

Medium 6

High 12

Very high 12


Fuzzy Logic

Input data values are rescaled using the assumption of

continuous membership values (i.e., fuzzyfication).

Environmental data are normally modeled using the symmetric

fuzzy models as generated by Kandel (1986):

21/1)( bxdxAFPx


Fuzzy Logic

Fuzzy operators allow the combination of layers containing fuzzy

values through a process of fuzzy overlay.

Operator Fuzzy Gamma:

yycombinação PAFSAF 1*

icombinaçãoSAF 11

icombinaçãoPAF


Fuzzy Logic

Operator Fuzzy Convex Sum.

If A1,.....,Ak are subsets of X, and w1,......,wk are non negative

weights then the convex combination of A1,....,Ak is:

AjjA w


Neural Networks

Clustering algorithms of the machine learning field.

Models of biological neurons and networks.

Unsupervised clustering

Self Organizing Maps (with and without k-means)

Fuzzy ArtMap


Neural Networks

Parameter SOM (without k-means) SOM (with k-means)

Input layer neurons 12 12

Output layer neurons 9 36

Initial neighborhood radius 5.24 9.49

Minimum learning rate 0.5 0.5

Maximum learning rate 1 1

Iterations 874,080 628,992

Quantization Error 0.0241 0.0187

SOM neural network parameters:


Neural Networks

Parameter Fuzzy ArtMap

F1 layer neurons 24

F2 layer neurons 6

Choice parameter 0.01

Learning rate 1

Vigilance parameter 0.95

Iterations 48,923

Fuzzy ArtMap neural network parameters:


Results and Discussion

Weighted overlay x Fuzzy logic:


Results and Discussion

Weighted overlay x Neural networks:



Conclusions and Future Studies

The evaluated methods are less intuitive, dependent on a number of

arbitrary parameters, demand more computational power, and do not

provide significant improvements when compared to the map produced

using weighted overlay,

Fuzzy logic seems to be a promising approach and further research will

be carried out in order to test different fuzzification methods, as well as

different fuzzy operators,

Neural networks will be disregarded due to the difficulties in setting the

necessary parameters, and

A framework to collect field data will be developed to provide a robust

base to carry out vulnerability map comparisons

Thank You !

Contact: [email protected]

Spatial Inference of Vegetation Vulnerability for the Ecological Economical Zoning of Minas Gerais Luis M. T. Carvalho 1 Moisés S. Ribeiro 2, Luciano T.

Documents

state of minas gerais

government of minas

high12 slide

fuzzy artmap slide

biologia slide

conservation priority

conservation degree

spatial heterogeneity